SVM Classifiers 18C 25x
Uriah Daugaard
1/2/2021
1 Info
Here I train the 16x video classifiers with the merged data, i.e. using stock culture data from every time point and light setting recorded.
2 Dependencies
library(tidyverse)
library(randomForest)
library(readr)
library(viridis)
library(e1071)
library(caret)
library(parallel)
library(doParallel)3 Data: load in and cleaning
3.1 Normal light
3.1.1 1st classifier data (2020-2021)
load("../../Class_18C_normalLight/1st_class_2020/data/25x/Coleps_irchel/Morph_mvt.RData")
dd25 <- morph_mvt
load("../../Class_18C_normalLight/1st_class_2020/data/25x/Colpidium/Morph_mvt.RData")
morph_mvt$species <- "Colpidium"
dd25 <- rbind(dd25, morph_mvt)
load("../../Class_18C_normalLight/1st_class_2020/data/25x/Euplotes/Morph_mvt.RData")
# morph_mvt <- morph_mvt %>% filter(mean_area>3000)
dd25 <- rbind(dd25, morph_mvt)
load("../../Class_18C_normalLight/1st_class_2020/data/25x/Euplotes_second/Morph_mvt.RData")
# morph_mvt <- morph_mvt %>% filter(mean_area>3000)
dd25 <- rbind(dd25, morph_mvt)
load("../../Class_18C_normalLight/1st_class_2020/data/25x/Paramecium_bursaria/Morph_mvt.RData")
dd25 <- rbind(dd25, morph_mvt)
load("../../Class_18C_normalLight/1st_class_2020/data/25x/Paramecium_caudatum/Morph_mvt.RData")
dd25 <- rbind(dd25, morph_mvt)
load("../../Class_18C_normalLight/1st_class_2020/data/25x/Stylonychia_1/Morph_mvt.RData")
morph_mvt$species <- "Stylonychia1"
dd25 <- rbind(dd25, morph_mvt)
load("../../Class_18C_normalLight/1st_class_2020/data/25x/Stylonychia_1_second/Morph_mvt.RData")
morph_mvt$species <- "Stylonychia1"
dd25 <- rbind(dd25, morph_mvt)
load("../../Class_18C_normalLight/1st_class_2020/data/25x/Stylonychia_2/Morph_mvt.RData")
morph_mvt$species <- "Stylonychia2"
dd25 <- rbind(dd25, morph_mvt)
load("../../Class_18C_normalLight/1st_class_2020/data/25x/Dexiostoma/Morph_mvt.RData")
dd25 <- rbind(dd25, morph_mvt)
load("../../Class_18C_normalLight/1st_class_2020/data/25x/Loxocephallus/Morph_mvt.RData")
dd25 <- rbind(dd25, morph_mvt)
load("../../Class_18C_normalLight/1st_class_2020/data/25x/Tetrahymena/Morph_mvt.RData")
morph_mvt$species <- "Tetrahymena"
dd25 <- rbind(dd25, morph_mvt)
load("../../Class_18C_normalLight/1st_class_2020/data/25x/Cryptomonas/Morph_mvt.RData")
morph_mvt$species <- "Cryptomonas"
dd25 <- rbind(dd25, morph_mvt)
load("../../Class_18C_normalLight/1st_class_2020/data/25x/Debris/Morph_mvt.RData")
morph_mvt$species <- "Debris_and_other"
morph_mvt <- morph_mvt %>% filter(mean_area<700)
dd25 <- full_join(dd25, morph_mvt)
# filtering
dd25 <- dd25 %>%
filter(net_disp>50, net_speed>5, N_frames>10)%>%
select(-edible_algae,-microcosm.nr)
dd25$magnification <- 2.5
# table(dd25$species)3.1.2 2nd classifier data (2022 Feb) - normal light
load("../../Class_18C_normalLight/2nd_class_2022Feb/data/25x/Coleps_irchel/Morph_mvt.RData")
dd25_2022feb <- morph_mvt
load("../../Class_18C_normalLight/2nd_class_2022Feb/data/25x/Colpidium/Morph_mvt.RData")
morph_mvt$species <- "Colpidium"
dd25_2022feb <- rbind(dd25_2022feb, morph_mvt)
load("../../Class_18C_normalLight/2nd_class_2022Feb/data/25x/Dexiostoma/Morph_mvt.RData")
dd25_2022feb <- rbind(dd25_2022feb, morph_mvt)
load("../../Class_18C_normalLight/2nd_class_2022Feb/data/25x/Euplotes/Morph_mvt.RData")
dd25_2022feb <- rbind(dd25_2022feb, morph_mvt)
load("../../Class_18C_normalLight/2nd_class_2022Feb/data/25x/Loxochephalus//Morph_mvt.RData")
dd25_2022feb <- rbind(dd25_2022feb, morph_mvt)
load("../../Class_18C_normalLight/2nd_class_2022Feb/data/25x/Paramecium_bursaria/Morph_mvt.RData")
# morph_mvt %>%
# ggplot(aes(mean_area))+
# geom_histogram()+
# geom_vline(xintercept = 3000)
morph_mvt <- morph_mvt %>% dplyr::filter(mean_area>3000)
dd25_2022feb <- rbind(dd25_2022feb, morph_mvt)
load("../../Class_18C_normalLight/2nd_class_2022Feb/data/25x/Paramecium_caudatum/Morph_mvt.RData")
dd25_2022feb <- rbind(dd25_2022feb, morph_mvt)
load("../../Class_18C_normalLight/2nd_class_2022Feb/data/25x/Stylonychia2_batch1/Morph_mvt.RData")
morph_mvt$species <- "Stylonychia2_batch1"
dd25_2022feb <- rbind(dd25_2022feb, morph_mvt)
load("../../Class_18C_normalLight/2nd_class_2022Feb/data/25x/Stylonychia2_batch2/Morph_mvt.RData")
morph_mvt$species <- "Stylonychia2_batch2"
dd25_2022feb <- rbind(dd25_2022feb, morph_mvt)
load("../../Class_18C_normalLight/2nd_class_2022Feb/data/25x/Tetrahymena/Morph_mvt.RData")
dd25_2022feb <- rbind(dd25_2022feb, morph_mvt)
# filtering
dd25_2022feb <- dd25_2022feb %>%
filter(net_disp>50, net_speed>5, N_frames>10)
dd25_2022feb$magnification <- 2.5
# table(dd25_2022feb$species)3.1.3 3rd data (20220301) - normal light
load("../../Class_18C_normalLight/3rd_data_20220301/data/25x/large_ciliates/Morph_mvt.RData")
dd25_20220301 <- morph_mvt
load("../../Class_18C_normalLight/3rd_data_20220301/data/25x/small_ciliates/Morph_mvt.RData")
dd25_20220301 <- rbind(dd25_20220301, morph_mvt)
# filtering
dd25_20220301 <- dd25_20220301 %>%
filter(net_disp>50, net_speed>5, N_frames>10)
dd25_20220301$magnification <- 2.5
# dd25_20220301 %>%
# ggplot(aes(mean_area))+
# geom_histogram()+
# facet_wrap(~species, scales = "free")
dd25_20220301 <- dd25_20220301 %>%
dplyr::filter(!(species %in% c("Paramecium_caudatum","Paramecium_bursaria","Euplotes") & mean_area<2000))
# table(dd25_20220301$species)Paramecium_caudatum maybe needs more cleaning
3.1.4 4th data (20220316) - normal light
load("../../Class_18C_normalLight/4th_data_20220316/data/25x/large_ciliates/Morph_mvt.RData")
dd25_20220316 <- morph_mvt
load("../../Class_18C_normalLight/4th_data_20220316/data/25x/small_ciliates/Morph_mvt.RData")
dd25_20220316 <- rbind(dd25_20220316, morph_mvt)
# filtering
dd25_20220316 <- dd25_20220316 %>%
filter(net_disp>50, net_speed>5, N_frames>10)
dd25_20220316$magnification <- 2.5
# dd25_20220316 %>%
# ggplot(aes(mean_area))+
# geom_histogram()+
# facet_wrap(~species, scales = "free")
dd25_20220316 <- dd25_20220316 %>%
dplyr::filter(!(species %in% c("Paramecium_caudatum","Paramecium_bursaria","Euplotes","Stylonychia_2") & mean_area<2000))
# table(dd25_20220316$species)3.1.5 5th data (20220406) - normal light
load("../../Class_18C_normalLight/5th_data_20220406/data/25x/large_ciliates/Morph_mvt.RData")
dd25_20220406 <- morph_mvt
load("../../Class_18C_normalLight/5th_data_20220406/data/25x/small_ciliates/Morph_mvt.RData")
dd25_20220406 <- rbind(dd25_20220406, morph_mvt)
# filtering
dd25_20220406 <- dd25_20220406 %>%
filter(net_disp>50, net_speed>5, N_frames>10)
dd25_20220406$magnification <- 2.5
# dd25_20220406 %>%
# ggplot(aes(mean_area))+
# geom_histogram()+
# facet_wrap(~species, scales = "free")
dd25_20220406 <- dd25_20220406 %>%
dplyr::filter(!(species %in% c("Paramecium_caudatum","Paramecium_bursaria","Euplotes","Stylonychia_2") & mean_area<2000))
# table(dd25_20220406$species)3.1.6 6th data (20220706) - normal light
load("../../Class_18C_normalLight/6th_data_20220706/data/25x/large_ciliates/Morph_mvt.RData")
dd25_20220706 <- morph_mvt
load("../../Class_18C_normalLight/6th_data_20220706/data/25x/small_ciliates/Morph_mvt.RData")
dd25_20220706 <- rbind(dd25_20220706, morph_mvt)
# filtering
dd25_20220706 <- dd25_20220706 %>%
filter(net_disp>50, net_speed>5, N_frames>10)
dd25_20220706$magnification <- 2.5
# dd25_20220706 %>%
# ggplot(aes(mean_area))+
# geom_histogram()+
# facet_wrap(~species, scales = "free")
dd25_20220706 <- dd25_20220706 %>%
dplyr::filter(!(species %in% c("Paramecium_caudatum","Paramecium_bursaria","Euplotes","Stylonychia_2") & mean_area<2000))
# table(dd25_20220706$species)3.2 Decreasing light
3.2.1 2nd classifier data (2022 Feb) - 18 perc
load("../../Class_18C_decreasingLight/Light_18perc/data/25x/Euplotes/Morph_mvt.RData")
dd25_2022feb_decrease <- morph_mvt
load("../../Class_18C_decreasingLight/Light_18perc/data/25x/Paramecium_bursaria/Morph_mvt.RData")
a <- morph_mvt %>%
ggplot(aes(mean_area))+
geom_histogram()
Pburs_dark_25x <- morph_mvt %>%
dplyr::select(mean_area,sd_area,mean_perimeter,sd_perimeter,mean_major,
sd_major,mean_ar,sd_ar,
mean_turning,sd_turning,gross_disp,max_net,net_disp,net_speed,max_step,
min_step,sd_step,sd_gross_speed,max_gross_speed,min_gross_speed)
set.seed(23)
Pburs_dark_25x_clust <- kmeans(Pburs_dark_25x, centers = 2, nstart = 25, iter.max = 1000)
Pburs_dark_25x$cluster <- as.factor(Pburs_dark_25x_clust$cluster)
b <- Pburs_dark_25x %>%
ggplot(aes(mean_area, col=cluster, fill=cluster))+
geom_histogram(position = "identity", alpha=0.3)
Pburs_dark_25x_clust <- kmeans(Pburs_dark_25x, centers = 3, nstart = 25, iter.max = 1000)
Pburs_dark_25x$cluster <- as.factor(Pburs_dark_25x_clust$cluster)
c <- Pburs_dark_25x %>%
ggplot(aes(mean_area, col=cluster, fill=cluster))+
geom_histogram(position = "identity", alpha=0.3)
Pburs_dark_25x_clust <- kmeans(Pburs_dark_25x, centers = 4, nstart = 25, iter.max = 1000)
Pburs_dark_25x$cluster <- as.factor(Pburs_dark_25x_clust$cluster)
d <- Pburs_dark_25x %>%
ggplot(aes(mean_area, col=cluster, fill=cluster))+
geom_histogram(position = "identity", alpha=0.3)
morph_mvt <- morph_mvt %>%
dplyr::mutate(cluster=as.factor(Pburs_dark_25x_clust$cluster)) %>%
dplyr::filter(cluster !="4") %>%
dplyr::select(-cluster)
e <- morph_mvt %>%
ggplot(aes(mean_area))+
geom_histogram()
# a + b + c + d + e + plot_annotation(title = "Cleaning of P.burs dark 25x")
dd25_2022feb_decrease <- rbind(dd25_2022feb_decrease, morph_mvt)
# filtering
dd25_2022feb_decrease <- dd25_2022feb_decrease %>%
filter(net_disp>50, net_speed>5, N_frames>10)
dd25_2022feb_decrease$magnification <- 2.5
# table(dd25_2022feb_decrease$species)3.2.2 3rd data (20220301) - 10 perc
load("../../Class_18C_decreasingLight/Light_10perc/data/25x/large_ciliates/Morph_mvt.RData")
dd25_20220301_decrease <- morph_mvt
load("../../Class_18C_decreasingLight/Light_10perc/data/25x/small_ciliates/Morph_mvt.RData")
dd25_20220301_decrease <- rbind(dd25_20220301_decrease, morph_mvt)
# filtering
dd25_20220301_decrease <- dd25_20220301_decrease %>%
filter(net_disp>50, net_speed>5, N_frames>10)
dd25_20220301_decrease$magnification <- 2.5
# dd25_20220301_decrease %>%
# ggplot(aes(mean_area))+
# geom_histogram()+
# facet_wrap(~species, scales = "free")
dd25_20220301_decrease <- dd25_20220301_decrease %>%
dplyr::filter(!(species %in% c("Paramecium_bursaria","Euplotes","Paramecium_caudatum") & mean_area<2500),
!(species %in% c("Coleps_irchel") & mean_area<1000))
# table(dd25_20220301_decrease$species)3.2.3 4th data (20220316) - 6 perc
load("../../Class_18C_decreasingLight/Light_6perc/data/25x/large_ciliates/Morph_mvt.RData")
dd25_20220316_decrease <- morph_mvt
load("../../Class_18C_decreasingLight/Light_6perc/data/25x/small_ciliates/Morph_mvt.RData")
dd25_20220316_decrease <- rbind(dd25_20220316_decrease, morph_mvt)
# filtering
dd25_20220316_decrease <- dd25_20220316_decrease %>%
filter(net_disp>50, net_speed>5, N_frames>10)
dd25_20220316_decrease$magnification <- 2.5
# dd25_20220316_decrease %>%
# ggplot(aes(mean_area))+
# geom_histogram()+
# facet_wrap(~species, scales = "free")
dd25_20220316_decrease <- dd25_20220316_decrease %>%
dplyr::filter(!(species %in% c("Paramecium_bursaria","Euplotes","Paramecium_caudatum") & mean_area<2000))
# table(dd25_20220316_decrease$species)3.2.4 5th data (20220406) - 1 perc
load("../../Class_18C_decreasingLight/Light_1perc/data/25x/large_ciliates/Morph_mvt.RData")
dd25_20220406_decrease <- morph_mvt
load("../../Class_18C_decreasingLight/Light_1perc/data/25x/small_ciliates/Morph_mvt.RData")
dd25_20220406_decrease <- rbind(dd25_20220406_decrease, morph_mvt)
# filtering
dd25_20220406_decrease <- dd25_20220406_decrease %>%
filter(net_disp>50, net_speed>5, N_frames>10)
dd25_20220406_decrease$magnification <- 2.5
# dd25_20220406_decrease %>%
# ggplot(aes(mean_area))+
# geom_histogram()+
# facet_wrap(~species, scales = "free")
dd25_20220406_decrease <- dd25_20220406_decrease %>%
dplyr::filter(!(species %in% c("Paramecium_bursaria","Euplotes","Paramecium_caudatum") & mean_area<2000))
# table(dd25_20220406_decrease$species)3.2.5 6th data (20220706) - 1 perc
load("../../Class_18C_decreasingLight/Light_1perc/data_20220706/data/25x/large_ciliates/Morph_mvt.RData")
dd25_20220706_decrease <- morph_mvt
load("../../Class_18C_decreasingLight/Light_1perc/data_20220706/data/25x/small_ciliates/Morph_mvt.RData")
dd25_20220706_decrease <- rbind(dd25_20220706_decrease, morph_mvt)
# filtering
dd25_20220706_decrease <- dd25_20220706_decrease %>%
filter(net_disp>50, net_speed>5, N_frames>10)
dd25_20220706_decrease$magnification <- 2.5
# dd25_20220706_decrease %>%
# ggplot(aes(mean_area))+
# geom_histogram()+
# facet_wrap(~species, scales = "free")
dd25_20220706_decrease <- dd25_20220706_decrease %>%
dplyr::filter(!(species %in% c("Paramecium_bursaria","Euplotes","Paramecium_caudatum") & mean_area<2000))
# table(dd25_20220706_decrease$species)4 Further data Processing
4.1 Merge data
dd25$data <- "Late 2020"
dd25_2022feb$data <- "20220201"
dd25_2022feb_decrease$data <- "20220201"
dd25_20220301$data <- "20220301"
dd25_20220301_decrease$data <- "20220301"
dd25_20220316$data <- "20220316"
dd25_20220316_decrease$data <- "20220316"
dd25_20220406$data <- "20220406"
dd25_20220406_decrease$data <- "20220406"
dd25_20220706$data <- "20220706"
dd25_20220706_decrease$data <- "20220706"
dd_30perc <- rbind(dd25,
dd25_2022feb,dd25_20220301 %>% dplyr::select(-Light_dark),
dd25_20220316 %>% dplyr::select(-Light_dark),
dd25_20220406 %>% dplyr::select(-Light_dark),
dd25_20220706 %>% dplyr::select(-Light_dark))
dd_18perc <- dd25_2022feb_decrease
dd_10perc <- dd25_20220301_decrease %>%
dplyr::select(-Light_dark)
dd_6perc <- dd25_20220316_decrease %>%
dplyr::select(-Light_dark)
dd_1perc <- rbind(dd25_20220406_decrease %>% dplyr::select(-Light_dark),
dd25_20220706_decrease %>% dplyr::select(-Light_dark))
dd_30perc$light <- "30%"
dd_18perc$light <- "18%"
dd_10perc$light <- "10%"
dd_6perc$light <- "6%"
dd_1perc$light <- "1%"
dd <- rbind(dd_30perc,dd_18perc,dd_10perc,dd_6perc,dd_1perc)
dd$species <- ifelse(dd$species == "Stylonychia_2", "Stylonychia2", dd$species)
dd$species <- ifelse(dd$species %in% c("Stylonychia2_batch1","Stylonychia2_batch2"), "Stylonychia2", dd$species)
# table(dd$species, dd$data, dd$magnification, dd$light)4.2 Filtering out outliers
If an individual is an outlier in more than 3 traits it gets excluded (about 7% gets excluded). Outlier based on boxplot definition.
dd$id <- 1:nrow(dd)
dd <- dd %>% na.omit()
dd_long <- dd %>%
dplyr::select(species, mean_area,sd_area,mean_perimeter,sd_perimeter,mean_major,
sd_major,mean_ar,sd_ar,
mean_turning,sd_turning,gross_disp,max_net,net_disp,net_speed,max_step,
min_step,sd_step,sd_gross_speed,max_gross_speed,min_gross_speed, id,
data, light, magnification) %>%
pivot_longer(cols=-c(id,species,data,light,magnification), names_to="variable") %>%
dplyr::group_by(variable,species,data,light,magnification) %>%
mutate(iqr = IQR(value),
first_quartile = quantile(value, probs = 0.25),
third_quartile = quantile(value, probs = 0.75),
outlier = ifelse(value<first_quartile-1.5*iqr | value>third_quartile+1.5*iqr,T,F))
outliers <- dd_long %>%
dplyr::filter(outlier==T) %>%
dplyr::group_by(species, id, data, light, magnification) %>%
dplyr::summarise(n = n()) %>%
dplyr::filter(n>3)
# table(outliers$species)
# nrow(outliers)/nrow(dd)
dd_filtered <- dd %>%
dplyr::filter(!is.element(id,outliers$id))
dd$removed_outliers <- F
dd_filtered$removed_outliers <- T
dd_comparison <- rbind(dd,dd_filtered)
dd <- dd_filtered
# dd_comparison %>%
# ggplot(aes((mean_area), col=removed_outliers))+
# geom_histogram(aes(fill=removed_outliers, y=..density..), position = "identity", alpha=0.3) +
# # geom_boxplot(outlier.alpha = 0.3) +
# theme_bw() +
# facet_wrap(species~interaction(light,data), scales = "free")
#
# dd_filtered %>%
# ggplot(aes(log10(mean_area)))+
# geom_density(aes(col=species))
# table(dd$species, dd$data, dd$magnification, dd$light)4.3 Sub-sampling to create training data
We have data from several time points and light settings. The goals is that in the final training data these are somewhat equally represented, regardless that the number of individuals tracked per species and time point can vary a lot (in other words: “even it out” across species, time point and light setting). This is important, as otherwise it can be that groups that are more abundant are weighted more.
dd <- dd %>%
mutate(data.light = interaction(data,light, drop = T),
data.light.species=interaction(data,light,species, drop = T))
print("number of individuals per species per date and light setting")## [1] "number of individuals per species per date and light setting"tab <- table(dd$data.light, dd$species)
tab##
## Coleps_irchel Colpidium Cryptomonas Debris_and_other Dexiostoma
## 20220406.1% 15 154 0 0 15
## 20220706.1% 6 178 0 0 203
## 20220301.10% 614 184 0 0 342
## 20220201.18% 0 0 0 0 0
## 20220201.30% 904 330 0 0 259
## 20220301.30% 587 193 0 0 317
## 20220316.30% 91 234 0 0 199
## 20220406.30% 73 93 0 0 100
## 20220706.30% 177 163 0 0 141
## Late 2020.30% 531 155 3475 623 649
## 20220316.6% 147 212 0 0 120
##
## Euplotes Loxocephallus Paramecium_bursaria Paramecium_caudatum
## 20220406.1% 35 362 118 137
## 20220706.1% 75 602 96 59
## 20220301.10% 33 1041 87 370
## 20220201.18% 708 0 1230 0
## 20220201.30% 228 967 488 445
## 20220301.30% 55 834 122 564
## 20220316.30% 136 687 860 786
## 20220406.30% 148 403 587 329
## 20220706.30% 411 653 1238 405
## Late 2020.30% 836 1090 1342 1017
## 20220316.6% 354 695 751 780
##
## Stylonychia1 Stylonychia2 Tetrahymena
## 20220406.1% 0 148 304
## 20220706.1% 0 18 1037
## 20220301.10% 0 448 1027
## 20220201.18% 0 0 0
## 20220201.30% 0 538 1035
## 20220301.30% 0 942 974
## 20220316.30% 0 642 1086
## 20220406.30% 0 625 554
## 20220706.30% 0 1592 968
## Late 2020.30% 345 146 1176
## 20220316.6% 0 469 676print("Sum of individuals per species")## [1] "Sum of individuals per species"colsums <- colSums(tab)
colsums## Coleps_irchel Colpidium Cryptomonas Debris_and_other
## 3145 1896 3475 623
## Dexiostoma Euplotes Loxocephallus Paramecium_bursaria
## 2345 3019 7334 6919
## Paramecium_caudatum Stylonychia1 Stylonychia2 Tetrahymena
## 4892 345 5568 8837Of course, Stylo1 is the one we the least individuals, because it went extinct and we do not have it anymore. As it went extinct, it shouldn’t be the limiting factor, I think. For now at least I’m taking 3 * Stylo1_sum as the number of individuals per species to be included (if a species/class has less than that all individuals are included).
Within a class I use that number so that roughly equally many individuals across time steps and light settings are included. As a last thing I divide the data into a training dataset and in a test dataset.
n <- min(colsums)*4
print(paste0("Max number of individuals used per species (if there are fewer for a species, then all are used): n=",n))## [1] "Max number of individuals used per species (if there are fewer for a species, then all are used): n=1380"num_ind_per_class <- dd %>% group_by(species) %>%
summarize(num_class = length(unique(data.light.species)),
num_ind_per_class = floor(n/num_class)) %>%
select(species,num_ind_per_class)
num_ind_per_class_vec <- c(num_ind_per_class$num_ind_per_class)
names(num_ind_per_class_vec) <- num_ind_per_class$species
set.seed(53)
split_up <- split(dd, f = dd$data.light.species)
split_up <- lapply(split_up, function(x) {
species <- unique(x$species)
x %>% sample_n(ifelse(nrow(x) < num_ind_per_class_vec[species], nrow(x), num_ind_per_class_vec[species]))})
trainingdata <- do.call("rbind", split_up)
print("Subsampled: number of individuals per species per date and light setting")## [1] "Subsampled: number of individuals per species per date and light setting"tab2 <- table(trainingdata$data.light, trainingdata$species)
tab2##
## Coleps_irchel Colpidium Cryptomonas Debris_and_other Dexiostoma
## 20220406.1% 15 138 0 0 15
## 20220706.1% 6 138 0 0 138
## 20220301.10% 138 138 0 0 138
## 20220201.18% 0 0 0 0 0
## 20220201.30% 138 138 0 0 138
## 20220301.30% 138 138 0 0 138
## 20220316.30% 91 138 0 0 138
## 20220406.30% 73 93 0 0 100
## 20220706.30% 138 138 0 0 138
## Late 2020.30% 138 138 1380 623 138
## 20220316.6% 138 138 0 0 120
##
## Euplotes Loxocephallus Paramecium_bursaria Paramecium_caudatum
## 20220406.1% 35 138 118 137
## 20220706.1% 75 138 96 59
## 20220301.10% 33 138 87 138
## 20220201.18% 125 0 125 0
## 20220201.30% 125 138 125 138
## 20220301.30% 55 138 122 138
## 20220316.30% 125 138 125 138
## 20220406.30% 125 138 125 138
## 20220706.30% 125 138 125 138
## Late 2020.30% 125 138 125 138
## 20220316.6% 125 138 125 138
##
## Stylonychia1 Stylonychia2 Tetrahymena
## 20220406.1% 0 138 138
## 20220706.1% 0 18 138
## 20220301.10% 0 138 138
## 20220201.18% 0 0 0
## 20220201.30% 0 138 138
## 20220301.30% 0 138 138
## 20220316.30% 0 138 138
## 20220406.30% 0 138 138
## 20220706.30% 0 138 138
## Late 2020.30% 345 138 138
## 20220316.6% 0 138 138print("Subsampled: Sum of individuals per species")## [1] "Subsampled: Sum of individuals per species"colsums2 <- colSums(tab2)
colsums2## Coleps_irchel Colpidium Cryptomonas Debris_and_other
## 1013 1335 1380 623
## Dexiostoma Euplotes Loxocephallus Paramecium_bursaria
## 1201 1073 1380 1298
## Paramecium_caudatum Stylonychia1 Stylonychia2 Tetrahymena
## 1300 345 1260 1380trainingdata <- trainingdata[complete.cases(trainingdata),]
# A stratified random split of the data
idx_train <- createDataPartition(trainingdata$species,
p = 0.65, # percentage of data as training
list = FALSE)
testdata <- trainingdata[-idx_train,]
trainingdata <- trainingdata[idx_train,]5 Loading in species compositions
species <- unique(dd$species)
species <- species[!is.element(species,c("Debris_and_other","Cryptomonas"))]
compositions <- read_csv("../../compositions.csv")## Rows: 15 Columns: 20## ── Column specification ────────────────────────────────────────────────────────
## Delimiter: ","
## chr (1): composition
## dbl (19): richness, Chlamydomonas, Cryptomonas, Monoraphidium, Cosmarium, St...##
## ℹ Use `spec()` to retrieve the full column specification for this data.
## ℹ Specify the column types or set `show_col_types = FALSE` to quiet this message.comp_name <- compositions$composition
compositions <- compositions[,species]
compositions_list <- apply(compositions, 1, function(x){
idx <- which(x==1)
names(idx)
})
names(compositions_list) <- comp_name
## filterting out Stylo1
compositions_list <- lapply(compositions_list, function(x){
x[x!="Stylonychia1"]
})6 Classifiers
6.1 Classifier plotting functions
classifier_plot_svm <- function(table, combination.nr){
# cm <- classifier$confusion
cm <- table
ncol <- ncol(cm)
cm <- apply(cm, 1, function(x) x/sum(x))
cm_long <- data.frame(Predicted = factor(rep(rownames(cm),ncol), levels = rownames(cm)),
Truth = factor(rep(colnames(cm), each=ncol), levels = colnames(cm)),
Classification = as.vector(cm))
plot <- cm_long %>%
ggplot(aes(Predicted,Truth,fill=Classification))+
geom_tile() +
geom_text(aes(label = round(Classification, 3)), col="white") +
scale_x_discrete(position = "top") +
scale_y_discrete(limits = rev(levels(cm_long$Truth))) +
scale_fill_viridis(option = "D", end=.8, discrete=FALSE)+
theme(axis.text.x = element_text(angle = 90, hjust = 0))+
theme(legend.text = element_text(size=14),
axis.title = element_text(size=14),
title = element_text(size=16),
axis.text = element_text(size=13))+
labs(title=paste("PPV of",combination.nr),fill="PPV")
return(plot)
}
classifier_assessment_plot <- function(cf, combination.nr){
cf.df <- cf$byClass[,1:4] %>%
as.data.frame() %>%
mutate(Species = gsub("Class: ", "", rownames(cf$byClass[,1:5]))) %>%
rename("NPV" = "Neg Pred Value", "PPV" = "Pos Pred Value",
"Sens" = "Sensitivity", "Spec" = "Specificity") %>%
pivot_longer(cols = 1:4) %>%
mutate(col = ifelse(value>=0.9,"1",
ifelse(value<0.9 & value>=0.8,"2",
ifelse(value<0.8 & value>=0.7,"3","4"))))
plot <- cf.df %>%
ggplot(aes(name,Species,fill=col))+
geom_tile() +
geom_text(aes(label = round(value, 3)), col="white") +
scale_x_discrete(position = "top") +
scale_y_discrete(limits = rev(levels(as.factor(cf.df$Species)))) +
scale_fill_manual(values = c("#006837","#66bd63","#f46d43","#a50026"), breaks = c("1","2","3","4")) +
theme(legend.text = element_text(size=14),
title = element_text(size = 16),
axis.title = element_blank(),
axis.text = element_text(size=13),
legend.position = "none")+
labs(title=combination.nr, fill="")
return(plot)
}6.2 Training of classifiers
formula <- factor(species) ~ mean_area + sd_area + mean_perimeter + sd_perimeter + mean_major +
sd_major + mean_ar + sd_ar +
mean_turning + sd_turning + gross_disp + max_net + net_disp + net_speed + max_step +
min_step + sd_step + sd_gross_speed + max_gross_speed + min_gross_speed
set.seed(474)
# classifiers_18c_25x <- list()
# classifiers_18c_25x_plots <- list()
# classifiers_18c_25x_assess_plots <- list()
# classified_test_data <- list()
completeList <- mclapply(1:length(compositions_list), function(i){
sub_trainingdata <- trainingdata %>%
filter(species %in% c(compositions_list[[i]],"Cryptomonas","Debris_and_other"))
sub_testdata <- testdata %>%
filter(species %in% c(compositions_list[[i]],"Cryptomonas","Debris_and_other"))
sub_trainingdata$species <- factor(sub_trainingdata$species)
sub_testdata$species <- factor(sub_testdata$species)
obj <- tune(svm, formula, data = sub_trainingdata, kernel="radial",
ranges = list(gamma = 2^(-8:2), cost = 2^(1:10)), probability = TRUE,
tunecontrol = tune.control(sampling = "fix", best.model = T))
classifiers_18c_25x <- obj$best.model
cf <- caret::confusionMatrix(predict(obj$best.model, newdata=sub_testdata %>% select(-species)),
sub_testdata$species)
sub_testdata <- sub_testdata %>%
dplyr::mutate(predicted = predict(obj$best.model, newdata=sub_testdata %>% select(-species)),
correct = species==predicted)
sub_testdata_summ <- sub_testdata %>%
group_by(data, species, light, magnification, correct) %>%
summarize(n = n()) %>%
dplyr::mutate(perc_corr = n/sum(n),
composition = names(compositions_list)[i]) %>%
dplyr::filter(correct==T)
classified_test_data <- sub_testdata_summ
classifiers_18c_25x_plots <- classifier_plot_svm(table = cf$table,
combination.nr = names(compositions_list)[[i]])
classifiers_18c_25x_assess_plots <- classifier_assessment_plot(cf, names(compositions_list)[[i]])
list(classifiers_18c_25x, classified_test_data, classifiers_18c_25x_plots, classifiers_18c_25x_assess_plots)
}, mc.cores = detectCores()-3)
classifiers_18c_25x <- map(completeList, 1)
classified_test_data <- map(completeList, 2)
classifiers_18c_25x_plots <- map(completeList, 3)
classifiers_18c_25x_assess_plots <- map(completeList, 4)
classified_test_data <- do.call("rbind", classified_test_data)
names(classifiers_18c_25x_plots) <- names(classifiers_18c_25x) <-
names(classifiers_18c_25x_assess_plots) <- comp_name6.3 Assessing of classifiers
There are mainly 4 measures:
Sensitivity: the probability that an individuals is classified as species X given that the it is of species X: P(Classified as species X | is of species X)
Specificity: the probability that an individuals is not classified as species X given that it is not of species X: P(Not classified as species X | is not of species X)
Positive Predictive Value PPV: the probability that an individual is of species X given that it has been classified as species X: P(is of species X | is classified as species X)
Negative Predictive Value NPV: the probability that an individuals is not of species X given that it has not been classified as species X: P(is not of species X | has not been classified as species X)
PPV is the most important for us!
library(patchwork)
for(i in 1:length(classifiers_18c_25x_plots)){
print(classifiers_18c_25x_plots[[i]] + classifiers_18c_25x_assess_plots[[i]] +
plot_layout(widths = c(4,2)))
}
classified_test_data <- classified_test_data%>%
dplyr::mutate(data=ifelse(data=="Late 2020","2020late",data),
tot_n = n/perc_corr, data=factor(data, ordered = T,
levels=c("2020late","20220201","20220301","20220316","20220406","20220706")),
light_num = as.numeric(gsub("%","",light)))
# classified_test_data %>%
# dplyr::filter(!(species %in% c("Cryptomonas","Debris_and_other"))) %>%
# ggplot(aes(data, perc_corr, col=composition, shape=light, size=tot_n)) +
# geom_hline(yintercept = 0.8, col="red") +
# geom_hline(yintercept = 0.9, col="green")+
# geom_point(position = position_jitter(0.2))+
# facet_wrap(~species, nrow = 3)+
# theme_bw() +
# scale_size_continuous(breaks = c(1,10,20,30)) +
# labs(size="Sample size of evaluation data", y="Correct identification in percentage", x="Date of data collection",
# title="Species identification across date, light intensities and compositions")
classified_test_data %>%
dplyr::filter(!(species %in% c("Cryptomonas","Debris_and_other"))) %>%
ggplot(aes(light_num, perc_corr))+
geom_hline(yintercept = 0.8, col="red") +
geom_point(aes(fill=composition,size=tot_n),shape=21,col="black",alpha=0.3)+
facet_wrap(~species, nrow = 3)+
scale_size_continuous(breaks = c(1,10,20,30)) +
geom_smooth(method = "lm")+
theme_bw() +
labs(size="Sample size of evaluation data", y="Correct species identification in percentage", x="Light intensity in percentage",
title="Species identification across light intensities and compositions")## `geom_smooth()` using formula 'y ~ x'
# classified_test_data %>%
# dplyr::filter(!(species %in% c("Cryptomonas","Debris_and_other"))) %>%
# ggplot(aes(tot_n,perc_corr, col=species))+
# geom_point()+
# theme_bw()7 Save classifiers
saveRDS(classifiers_18c_25x, "svm_video_classifiers_18c_25x_20220706_MergedData.rds")
# cl <- readRDS("classifiers_18c_25x.rds")